Open ahnaf1393 opened 1 year ago
Hi!
I am getting unexpected results while training the unconditional generator. I was training using 2 RTX 3090 (24 GB) GPUs. Due to memory limitations, I changed the batch size to 8. I used the following command to train:
python run.py --dataset shapenet_cars --path_length_regularization --gpus 2 --batch_size 8
Additional training seems to decrease the quality of the predicted images. For example, this was the RGB prediction after 4999 iterations:
And this was after 94999:
The loss curves also seem to follow a similar pattern:
Is this all because of the small batch size or am I doing something wrong?
Hi!
I am getting unexpected results while training the unconditional generator. I was training using 2 RTX 3090 (24 GB) GPUs. Due to memory limitations, I changed the batch size to 8. I used the following command to train:
python run.py --dataset shapenet_cars --path_length_regularization --gpus 2 --batch_size 8
Additional training seems to decrease the quality of the predicted images. For example, this was the RGB prediction after 4999 iterations:
And this was after 94999:
The loss curves also seem to follow a similar pattern:
Is this all because of the small batch size or am I doing something wrong?